ai AM market analysis — 2026-10-19

The AI sector’s structural debate is shifting from who builds the largest frontier model to who controls the deployment layer. The UK’s Competition and Markets Authority notes that smaller, more efficient foundation models are widening where AI can run, including local inference on consumer hardware, while Apple is building its platform strategy around on-device models for app features, guided generation, and tool calling. That split points to two competing routes to value: centralised frontier systems kept behind cloud APIs, and leaner models embedded directly into devices and software.

Open foundation models sit awkwardly across that split. Stanford HAI argues they can broaden access and speed up innovation, but the same openness raises harder questions about transparency, misuse, and safety obligations that closed systems can sidestep more easily. The CMA’s analysis adds a structural caveat: regardless of how models are packaged, competition still concentrates around who controls compute, data, specialist expertise, funding, and the incumbent platforms that touch all four. On-device efficiency gains do not obviously loosen that grip so much as add a new front where it can be exercised.

Taken together, the source pack describes a market still forming around these tensions rather than one that has resolved them. The practical signal for this session is adoption and governance, not pricing: whether on-device models find real developer uptake, and whether open-model oversight keeps pace with their spread.